Introducing support for remote MCP servers, image generation, Code Interpreter, and more in the Responses API

Announcements
26

It is proprietary.

So …

27

Sorry if its been asked is streaming working for image gen in responses? When I try it all comes back at once and takes a minute despite the fact I set stream=true ? :frowning:

28

It is natural that no events are streamed while a tool call is processing.

For the image generation tool in particular, you can request up to three preview images to be returned during the generation, which can serve as both keep-alive and to give something to display about the ongoing work being done.

tools=[{"type": "image_generation", "partial_images": 2}],

Each preview costs an additional $0.004

1 Like
29

We would love to see MCP being used for LLM → business api communication, as an alternative for the browser automation path. What do you think?

31

What do you mean by “LLM → business api”?

32

Have you set the # of partial images? Streaming for images works a bit differently than text. You’ll need to define how many parts of the image (from top to bottom) should stream back, up to 3 parts.

stream: true,
tools: [{ type: "image_generation", partial_images: 3 }],
33

You can check this discussion!

Basically GUI Agents vs API Agents

34

Been leveraging the responses API to showcase work we’re doing with MCP servers. Great work peeps!

1 Like
35

I have a question.

I hope it’s ok to ask.

The image shows a section of an interface where GitHub is listed as an enabled connector in ChatGPT under the 'Enabled connectors' heading. (Captioned by AI)
The image shows a section of an interface where GitHub is listed as an enabled connector in ChatGPT under the 'Enabled connectors' heading. (Captioned by AI)590×269 5.64 KB

When can a single user leverage MCP outside a team?

I cant afford to pay for 2 users just to use MCP…

And what is the EU/UK deal here… OpenAI… Hold governments to account! Don’t just deny access… WHY can I not access this service technically? (Or politically?) What are the parameters?

@edwinarbus

If I can and my understanding is wrong please show me how?

1 Like
37

That’s ChatGPT. This Topic is about the Responses API

38

I’m trying to use the MCP feature of the Responses API.

The documentation does not explain whether there is a timeout for request duration (answer: yes, there is — but where is this documented?). Which features does the client support (e.g., progress notifications? Answer: no, they’re not working).

Okay, so if MCP is advertised as the de-facto choice for agent-to-agent collaboration and external systems integration, how can this work with real agentic systems and long operations (i.e., durations longer than 60 seconds)?

I find the current documentation severely lacking; it’s nothing more than “try and hope it works — if it doesn’t, too bad.” OpenAI’s support for developers is really frustrating. Marketing claims are not what we expect.